Loop as a Bridge: Can Looped Transformers Truly Link Representation Space and Natural Language Outputs?

TL;DR

This paper investigates whether Looped Transformers can connect internal representations with natural language outputs, finding that increased looping narrows the gap but degrades internal knowledge and lacks iterative perception improvements.

Contribution

It provides empirical evidence on the limitations of Looped Transformers in bridging internal representations and language outputs, highlighting issues with knowledge degradation and perception.

Findings

Loop iterations narrow the gap between internal knowledge and outputs

Increased loops degrade internal representations

Perception of representations does not improve across loops

Abstract

Large Language Models (LLMs) often exhibit a gap between their internal knowledge and their explicit linguistic outputs. In this report, we empirically investigate whether Looped Transformers (LTs)--architectures that increase computational depth by iterating shared layers--can bridge this gap by utilizing their iterative nature as a form of introspection. Our experiments reveal that while increasing loop iterations narrows the gap, it is partly driven by a degradation of their internal knowledge carried by representations. Moreover, another empirical analysis suggests that current LTs' ability to perceive representations does not improve across loops; it is only present in the final loop. These results suggest that while LTs offer a promising direction for scaling computational depth, they have yet to achieve the introspection required to truly link representation space and natural language.